Omnistat Documentation#
Welcome to the documentation area for Omnistat, a set of Python utilities and data collectors to support scale-out cluster telemetry targeting AMD Instinct™ MI accelerators. Omnistat operates in two primary modes: permanent, cluster-wide collection managed by a system administrator, and temporary collection scoped to a single user’s batch job. Collected telemetry can be visualized in Grafana, exported for custom analysis, or explored interactively by an AI coding agent.
Project overview, supported metrics, and advanced profiling with hardware counters and kernel tracing.
Cluster-wide deployment by a system administrator, plus Grafana integration for visualization.
Per-job collection without elevated credentials, including job execution and analysis.
Agent skills for exploring collected telemetry with an AI coding agent, and the developer guide for adding new data collection mechanisms.